The Detection of Colorectal Cancer through Machine Learning-Based Breath Sensor Analysis.

Area of Science:

• Oncology
• Biomedical Engineering
• Analytical Chemistry

Background:

• Colorectal cancer (CRC) is a leading cause of cancer death globally, necessitating improved screening methods.
• Current screening tools have limitations, driving research into novel, non-invasive diagnostic approaches.
• Volatile organic compounds (VOCs) in breath are emerging biomarkers for various diseases, including cancer.

Purpose of the Study:

• To evaluate the diagnostic potential of a table-top breath analyzer for detecting colorectal cancer (CRC).
• To assess the performance of different machine learning algorithms in analyzing breath VOCs for CRC identification.
• To demonstrate the feasibility of a point-of-care breath test for CRC screening.

Main Methods:

• Breath samples were collected from 105 patients with CRC and 186 non-cancer subjects.
• A table-top breath analyzer equipped with metal-oxide-semiconductor (MOX) sensors was utilized for sample analysis.
• Supervised machine learning algorithms, including Random Forest, C4.5, Artificial Neural Network, and Naïve Bayes, were employed for data analysis.

Main Results:

• Random Forest with Evolutionary Search for Features achieved 79.3% accuracy, 53.3% sensitivity, and 93.0% specificity (AUC ROC 0.734).
• Artificial Neural Networks with Greedy Search for Features yielded 78.2% accuracy, 43.3% sensitivity, and 96.5% specificity (AUC ROC 0.735).
• The MOX sensor combination effectively distinguished between healthy and CRC-affected individuals' breath samples.

Conclusions:

• The developed breath analyzer demonstrates significant potential as a tool for identifying and categorizing CRC in a clinical setting.
• The non-invasive, rapid, and targeted nature of this breath test offers promising prospects for future CRC screening applications.
• Breath analysis using sensor technology represents a viable and innovative approach for early cancer detection.